WO2010037314A1

WO2010037314A1 - A method for searching and the device and system thereof

Info

Publication number: WO2010037314A1
Application number: PCT/CN2009/073971
Authority: WO
Inventors: 胡汉强
Original assignee: 华为技术有限公司
Priority date: 2008-09-26
Filing date: 2009-09-16
Publication date: 2010-04-08
Also published as: US20110173192A1; EP2352102A1; EP2352102A4; CN101685456A; US8527509B2; CN101685456B

Abstract

A method for searching is provided which includes the following steps: receiving the searching request; extracting the user interest model from the user personalized data according to the searching request; obtaining the meta index of each member engine; selecting the member engine according to the meta index of each member engine, the searching request and the user interest model; sending the searching request to the selected member engine and the selected member engine completing searching. The system and relative device thereof are provided such as a searching server, a scheduling server. This improves the efficiency of system and precision of searching.

Description

Method, system and device for searching, the present application claims priority to Chinese patent application filed on September 26, 2008, the Chinese Patent Office, Application No. 200810216521.5, entitled "A Search Method, System and Apparatus" And the priority of the Chinese Patent Application entitled "Method, System and Apparatus for Searching", filed on December 24, 2008, filed on Dec. 24, 2008, the entire disclosure of which is incorporated herein by reference. In the application.

Technical field

The present invention relates to the field of communication technologies, and in particular, to a search method, system and apparatus.

Background of the invention

With the development and advancement of science, communication technology has also developed rapidly, and mobile search technology is a new bright spot along with the development of communication technology. Research on mobile search technology has also become the focus of industry research and development. A very important technical highlight of mobile search technology is accurate search, which is to provide users with personalized search services to achieve the user's search.

In a prior art processing scheme of meta-search, the search server collects the meta-index of the member engine, and the search server calculates the similarity between the search request and the member engine according to the meta-index, and selects a member engine with high similarity to serve the user, Search requests are distributed to these selected member engines for searching.

In the process of implementing the present invention, the inventors found that the selected member engine in the meta search scheme may be inaccurate, resulting in low accuracy of the search.

Summary of the invention

In order to improve the accuracy of the search, embodiments of the present invention provide corresponding methods, systems, and apparatus for searching. A method for searching, the method comprising: receiving a search request; extracting a user's interest model from the user personalized data according to the search request; acquiring a meta index of each member engine; and searching, according to the meta index of each member engine The request and the user's interest model select a member engine, and send the search request to the selected member engine to facilitate the selected member engine to complete the search.

Correspondingly, a system for searching, the system capable of applying the above search method, the system comprising: a search service subsystem, configured to receive a search request, and receive a meta index reported by each member engine, according to each member engine a meta index, the search request, and a user's interest model selection member engine, to send the search request to the selected member engine;

At least one member engine is configured to report the meta index of the member engine to the search service subsystem, and complete the search after receiving the search request sent by the search service subsystem.

Each of the foregoing embodiments extracts the interest model of the user after receiving the search request, and selects a member engine according to the meta index of the member engines, the search request, and the interest model of the user, that is, the selected member engines. Taking full account of the factors of the search request and the user's interest model, after the search is completed by the selected member engines, the selection of the member engine is more personalized, and the selected engine is related to the user's interest, thus improving the system scheduling (or Select) the efficiency and accuracy of the search.

BRIEF DESCRIPTION OF THE DRAWINGS

1 is a schematic structural diagram of a system according to an embodiment of the present invention; 2 is a schematic structural diagram of a system according to another embodiment of the present invention;

3 is a schematic structural view of a device in the system architecture shown in FIG. 2;

4 is a schematic structural diagram of a system according to another embodiment of the present invention;

Figure 5 is a schematic structural view of the device in the system architecture shown in Figure 4;

6 is a schematic structural diagram of a system according to another embodiment of the present invention;

7 is a schematic structural view of a device in the system architecture of FIG. 6;

8 is a schematic flow chart of an embodiment of a search method according to the present invention;

9 is a schematic flow chart of another embodiment of a search method according to the present invention;

10 is a schematic flow chart of another embodiment of a search method according to the present invention;

11 is a schematic flow chart of extracting a user interest model according to the present invention;

12 is a schematic flow chart of extracting a static interest model of a user according to the present invention;

13 is a schematic flow chart of extracting a dynamic interest model of a user according to the present invention;

14 is a schematic flow chart of a selected member engine according to the present invention;

15 is a schematic structural diagram of still another search system according to an embodiment of the present invention;

16 is a structural diagram of a search server and an application server in the search system architecture of FIG. 15;

Figure 17 is a flow chart showing the operation of the system of the search shown in Figure 15.

Mode for carrying out the invention

Referring to FIG. 1, a schematic diagram of a system structure of a system embodiment for searching includes:

Search client 10, used to send a search request to the search service subsystem,

The search service subsystem 20 is configured to receive a search request sent by the search client, extract the interest model of the user from the user personalized data according to the search request, obtain a meta index reported by each member engine, and obtain a meta element according to the member engine The index, the search request, and the user's interest model select a member engine; the search request is sent to the selected member engine;

At least one member engine for providing a meta index of the member engine to the search service subsystem, and completing the search after receiving the search request sent by the search service subsystem. In the implementation process, there are generally multiple member engines, such as a first member engine 301, a second member engine 302, a third member engine 303, and a fourth member engine 304.

It should be noted that, in various embodiments of the present invention, the member engine refers to each vertical search engine responsible for a specific search in the architecture of the meta search. The meta index refers to the statistical data used to describe the capabilities of the member engine and used for member engine selection in the metasearch architecture. Specifically, the member engine's meta index is a document or record contained in a database, a sub-database, a database, or a sub-database corresponding to the member engine, and statistics of terms contained in the document or the record. In various embodiments of the invention, the meta-index of the member engine is used as one of the basis for selecting a member engine.

Interest model is a vector of dimension with respect to the weight of a certain value consisting weights, it can be expressed using the formula R = (rl, r2, .. ri .... , rn); wherein _n represents _{the n} dimensions, 'is The weight score of the i-th dimension of the interest model, and i is a natural number. The user's interest model is based on the weighted scores of certain dimensions extracted from the relevant data of the user. In the above system for searching, since the search service subsystem 20 extracts the interest model of the user after receiving the search request, and selects the member engine according to the meta index of the respective member engines, the search request, and the interest model of the user, That is to say, the member engines of these choices fully consider the factors of the search request and the user's interest model. After that, the selected member engines complete the search, the member engine selection is more personalized, and the selected engine is interested in the user. Relatedly, this improves the efficiency of the system scheduling (or selection) and the accuracy of the search.

Referring to Figure 2, there is shown a block diagram of another embodiment of a search system similar to that described above with respect to Figure 1, including a search client 10, a search service subsystem 20A, and at least one member engine. The search client 10 is configured to send a search request to the search service subsystem; at least one member engine is configured to provide a meta index of the member engine to the search service subsystem, and after receiving the search request sent by the search service subsystem, Finish the search. In the implementation process, there are generally multiple member engines, such as a first member engine 301, a second member engine 302, a third member engine 303, and a fourth member engine 304.

The search service subsystem 20A is comprised of a search server 201 A and a user database 202.

The user database 202 is configured to store or provide personalized data of the user;

The search server 201A is configured to receive a search request sent by the search client, extract the interest model of the user from the user personalized data according to the search request, obtain a meta index reported by each member engine, and obtain a meta element according to each member engine. The index, the search request, and the user's interest model select the member engine; the search request is sent to the selected member engine.

Referring to FIG. 3, a detailed structural diagram of each device in the system architecture shown in FIG. 2, wherein the search server 201A includes:

The search request receiving module 201A1 is configured to receive a search request sent by the search client;

The user's interest model extraction module 201A2 extracts the user's interest model from the user personalized data according to the search request;

a meta index collection module 201A3, configured to obtain a meta index of each member engine;

a member engine selection module 201A4, configured to select a member engine according to the meta index of the respective member engines, the search request, and the interest model of the user;

The search request distribution module 201A5 is configured to send the search request to the selected member engine, so that the selected member engine completes the search according to the search request.

Referring to FIG. 4, a schematic diagram of an architecture of another search system embodiment, similar to the system of FIG. 1, includes a search client 10, a search service subsystem 20B, and at least one member engine. The search client 10 is configured to send a search request to the search service subsystem; at least one member engine is configured to provide a meta index of the member engine to the search service subsystem, and a part of the member engines receive the send by the search service subsystem After searching for the request, complete the search. In the course of implementation, there are typically a plurality of member engines, such as a first member engine 301, a second member engine 302, a third member engine 303, and a fourth member engine 304.

The search service subsystem 20B is composed of a search server 201B, a dispatch server 203A, and a user database 202.

Specifically, the user database 202 is configured to store or provide personalized data of the user;

The search server 201B is configured to receive a search request sent by the search client, extract the interest model of the user from the user personalized data according to the search request, and send the user's interest model and the search request to the dispatch server 203A. Receiving the selected member engine returned by the dispatch server 203A, and transmitting the search request to the selected member engine; The scheduling server 203A is configured to receive the interest model of the user sent by the search server 201B and the search request, and receive a meta index reported by each member engine; according to the meta index of the member engines, the search request, and the interest of the user The model selects a member engine; the selected member engine is returned to the search server 201B.

Referring to FIG. 5, it is a schematic structural diagram of the search server 201B and the dispatch server 203A in the system architecture shown in FIG.

The search server 201B includes:

The search request receiving module 201B1 is configured to receive a search request sent by the search client, and send the search request to the member engine selection request sending module 201B3;

The user's interest model extraction module 201B2 is configured to extract the user's interest model from the user personalized data according to the search request, and send the user's interest model to the member engine selection request sending module 201B3;

a member engine selection request sending module 201B3, configured to send the search request and the user's interest model to a scheduling server, so that the scheduling server is configured according to the meta index of the respective member engines, the search request, and the user The interest model selects the member engine;

a member engine selection result receiving module 201B4, configured to receive a selected member engine returned by the scheduling server; a search request distribution module 201B5, configured to send the search request to the selected member engine, to facilitate the selection The member engine completes the search based on the search request.

The dispatch server 203A that communicates with the search server 201B described above includes:

a member engine selection request receiving module 203A1, configured to receive a search request sent by the search server and an interest model of the user;

a meta index collection module 203A2, configured to obtain a meta index of each member engine;

a member engine selection module 203 A3, configured to select a member engine according to the meta index of the respective member engines, the search request, and the interest model of the user;

a member engine selection result returning module 203A4, configured to send the selected member engine to the search server, so that the search server sends the search request to the selected member engine, the selected member engine The search is completed according to the search request.

Referring to FIG. 6, a schematic diagram of an architecture of another search system embodiment, similar to the system described above with reference to FIG. 1, includes a search client 10, a search service subsystem 20C, and at least one member engine. The search client 10 is configured to send a search request to the search service subsystem; at least one member engine is configured to provide a meta index of the member engine to the search service subsystem, and after receiving the search request sent by the search service subsystem, Finish the search.

The search service subsystem 20C shown in FIG. 6 is composed of a search server 201C, a dispatch server 203C, and a user database 202, wherein

The search server 201C is configured to receive a search request sent by the search client, send the search request to the dispatch server, receive the selected member engine returned by the dispatch server, and send the search request to the selected member engine. The scheduling server 203C is configured to receive the search request sent by the search server, and personalize the user according to the search request. Extracting the user's interest model from the data, and obtaining a meta index of each member engine; selecting a member engine according to the meta index of the respective member engine, the search request, and the user's interest model; returning the selected member engine to the Search server.

Referring to FIG. 7, the structure of the search server 201C and the dispatch server 203C is shown in the system architecture shown in FIG. 6. The search server 201C as shown in Fig. 7 includes:

The search request receiving module 201C1 is configured to receive a search request sent by the search client, and send the search request to the member engine selection request sending module 201C3;

a member engine selection request sending module 201C3, configured to send the search request to the scheduling server, so that the scheduling server selects a member engine according to the meta index of the respective member engines, the search request, and the user's interest model;

a member engine selection result receiving module 201C4, configured to receive a selected member engine returned by the scheduling server; a search request distribution module 201C5, configured to send the search request to the selected member engine, to facilitate the selection The member engine completes the search based on the search request.

As shown in FIG. 7, a dispatch server 203C, which can communicate with the aforementioned search server 201C, includes

a member engine selection request receiving module 203C1, configured to receive a search request sent by the search server;

The user's interest model extraction module 203C5 is configured to extract the user's interest model from the user personalized data according to the received search request, and send the user's interest model to the member engine selection module 203C3;

a meta index collection module 203C2, configured to obtain a meta index of each member engine;

a member engine selection module 203C3, configured to select a member engine according to a meta index of each member engine, a search request, and a user's interest model;

a member engine selection result returning module 203C4, configured to send to the search server according to the selected member engine, so that the search server sends the search request to the selected member engine, the selected member engine The search is completed according to the search request.

The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, ie may be located One place, or it can be distributed to multiple network elements. Some or all of the modules may be selected according to actual needs to achieve the objectives of the embodiments of the present invention. Those of ordinary skill in the art can understand and implement without deliberate labor.

The above-mentioned system and device for searching, because the member engine is selected according to the meta index of each member engine, the search request, and the user's interest model, and the search is performed by the selected member engines, thereby further improving the accuracy of the search and improving the search. effectiveness.

The aforementioned workflows for searching systems and search servers, scheduling servers, etc. are similar to the search methods below, as will be described in more detail below.

Referring to FIG. 8, a flow chart of a search method of an embodiment is shown. The method includes:

801. The search service subsystem receives a meta index of each member engine.

Specifically, each member engine may report its meta index to the search service subsystem. Each member engine corresponds to a database, and the database may include several sub-databases. The meta-index of the member engine is specifically determined by the data in the database or sub-database corresponding to the member engine. The meta index contains statistics about terms in the data or interest in documents in the data Model statistics. The interest model of the document is a vector based on the weighted scores of the data extracted in the document relative to a certain dimension. In a specific embodiment of the invention, the dimension of the interest model of the above document and the interest model of the user should be consistent.

More specifically, the meta index may include one of the following information or any combination thereof:

1. The term maximum normalized weight vector mnw=(mnwl, mnw2, , mnwi, mnwp), where mnwi is the maximum normalized weight of the term ti relative to all documents in the database or subdatabase corresponding to the member engine.

Where mnwi can be calculated in the following way: first calculate the weight of each document in the database relative to the term ti, the value of the weight is the number of occurrences of the term in the document, and then the maximum weight of all documents relative to the term ti Value, the maximum weight mnwi' of the term ti relative to all documents in the database is obtained, and the vectors mnw' mnwl', mnw2', ... mnwi', ...., mnwp') are normalized. , finally get the normalized vector mnw of mnw' vector (mnwl, mnw2, ..., mnwi, , mnwp

2. The term average normalized weight vector anw = (anwl, anw2, , anwi , anwp), where anwi is the average normalized weight of the term ti relative to all documents in the database or subdatabase corresponding to the member engine.

Among them, anwi can be calculated as follows: First, calculate the weight of each document of the database relative to the term ti, the value of the weight is the number of occurrences of the term in the document, and then average from the weight of all documents relative to the term ti, Obtain the average weight anwi' of the term ti relative to all documents in the database, and then normalize the vector anw' (anwl', anw2' ... anwi', .... . , anwp'), and finally Get the normalized vector anw of anw' vector (anwl, anw2,..., anwi, , anwp

3. The interest model of the document in the database or sub-database has the largest normalized weight vector mnv=(mn _V l, mnv2, ..., mnvi, ..., mnvn ), where mnvi is the The maximum normalized weight of the i-th dimension of the document's interest model relative to all documents in the database or sub-database corresponding to the member engine.

Among them, mnvi can be calculated in the following way: First, calculate the weight of each document of the database relative to the i-th dimension of the interest model. The value of the weight is the i-th dimension range of the interest model (such as: sports) in the document. The sum of the word frequencies of all words; and then taking the maximum value from the weights of all documents relative to the i-th dimension of the interest model, the maximum weight mnvi' of the i-th dimension of the interest model relative to all documents in the database D is obtained. The vector mnv'(mnvl', mnv2' ... mnvi', ....., mnvp') is normalized, and finally the normalized vector mnv of the mnv' vector is obtained (mnvl, mnv2,... , mnvi, , mnvp).

4. The interest model of the document in the database or subdatabase averages the normalized weight vector an _V = (an _V l, anv2, ..., anvi, ..., anvn), where anvi is The average normalized weight of the i-th dimension of the document's interest model relative to all documents in the database corresponding to the member engine.

Among them, anvi can be calculated in the following way: First, calculate the weight of each document in the database relative to the i-th dimension of the interest model. The value of the weight is the i-th dimension of the interest model in the document (such as: sports). The sum of the word frequencies of all words; then calculate the average from the weights of all the documents relative to the i-th dimension of the interest model, and get the average weight of the i-th dimension of the interest model relative to all documents in the database anvi', then The vector an _V '(an _V r,an _V 2' ...an _V i', .... . , anvp') is normalized, and finally the normalized vector anv of the anv' vector (anvl, Anv2,...,anvi, ,anvp).

5. The term ti is relative to the global reverse document frequency gidfi of the database, where gidf=l/dfi, dfi is the number of documents containing the term ti in the database corresponding to the meta index. 6. The global reverse document frequency IM _gidfi corresponding to the i-th dimension of the document's interest model, where IM _gidfi=l/IM_IDFi, IM-IDFi is the i-th dimension of the interest model belonging to the document contained in the database or sub-database The number of documents for one or more terms.

802. The search service subsystem receives a search request sent by the search client.

Generally, the search request carries the ID of the user, and information such as a search keyword composed of one or more terms. In the following text, β = (gl, g2, ..g ... g :) can be used to represent the vector corresponding to the search request, where ^ is the weight of the term in the search request, and i and k are natural numbers.

803. The search service subsystem extracts a user's interest model from the user database.

Specifically, the user database generally stores personalized data of the user, including the user's static user profile search history, presence information, location information, and the like. The user's interest model can be extracted by different methods for different data, which will be detailed later.

804. Select a member engine according to the search request, the user's interest model, and the meta index.

Specifically, the step of selecting considers both the search request and the meta-index, and more factors of the user's interest model, and the specific process will be described later.

805. Distribute the search request to the selected member engine. Referring to Figure 9, a flow diagram of another search method, the method includes:

901. The scheduling server receives each member engine element index.

Specifically, each member engine may report its meta index to the scheduling server.

902. The search server receives a search request sent by the search client.

Generally, the search request carries the ID of the user, and information such as a search keyword composed of one or more terms. It can be seen that the foregoing steps 901 and 902 may have no prioritized sequence.

903. After receiving the search request, the search server extracts the user's interest model from the user personalized data in the user database.

904. The search server submits the search request and the user's interest model to the scheduling server.

905. The scheduling server selects a member engine according to the search request, the user's interest model, and the meta index.

906. The scheduling server sends the selected member engine to the search server.

907. The search server sends the foregoing search request to the selected member engine, so that the member engine completes the search. Referring to Figure 10, for another similar search method, the method includes:

1001. The scheduling server receives a meta index of each member engine.

1002. The search client sends a search request to the search server.

Generally, the search request carries the ID of the user, and information such as a search keyword composed of one or more terms. It can be seen that the steps of 1001 and 1002 have no prioritization in time. 1003. The search server sends the search request to the scheduling server after receiving the search request.

1004. After receiving the search request, the dispatching server extracts a user's interest model from user personalized data in the user database.

1005. The scheduling server selects a member engine according to the search request, the user's interest model, and the meta index.

1006. The scheduling server returns the selected member engine to the search server.

1007. After receiving the selected member engine, the search server distributes the search request to the selected member engine. The specific method of extracting the user's interest model from the user personalized data in the user database in the foregoing steps 803, 903 or 1004 is specifically described below.

Referring to FIG. 11, a method for extracting the user's interest model from user personalized data includes:

1101, representing a user's interest with a number of interest dimensions;

The dimensions of interest such as: news, sports, entertainment, finance, technology, real estate, games, women, forums, weather, merchandise, home appliances, music, reading, blog, mobile, military, education, travel, MMS, ring tones, catering, civil aviation , industry, agriculture, computers, geography, etc. Of course, the actual implementation process is not limited to the above dimension of interest.

1102, giving a rating value for each interest dimension;

1103. Form, by the score value for each interest dimension, a vector, where the vector is an interest model of the user.

The user's interest model can be expressed as ^{/? =} ^ ¹ , ² , '^' - " ⁷ ), where n represents n dimensions, which is the weight score of the i-th dimension of the user's interest model, i is Specifically, the user personalized data may be a static user profile of the user, and the obtained user's interest model may be referred to as a user static interest model. Referring to FIG. 12, the method specifically includes:

1201. Obtain a word frequency of a word belonging to a certain interest dimension in a static user profile of the user;

1202. Calculate a sum of word frequencies of all words belonging to the interest dimension as a score value of the interest dimension;

1203. Each score value constitutes a score value vector, and the score value vector is a static interest model of the user.

The user's static interest model can be expressed as: The interest model corresponding to the user's static user profile Rl=(pl,

_P 2, p3 , ..., pi ), where pi is the sum of the word frequencies of all words belonging to the i-th interest dimension in the static user profile, and i is a natural number. Specifically, the user personalized data may also be a user's search history, and the obtained user's interest model may be referred to as a user's dynamic interest model. Referring to FIG. 13, the method may specifically include:

1301. Obtain a word frequency of a word belonging to a certain interest dimension in a document that is clicked in the search history of the user;

1302. Calculate a sum of word frequencies of all words in the document belonging to the interest dimension, and the sum of the word frequencies is used as a score value of the document for the interest dimension;

1303. Forming a score value vector for the document for each score value of each interest dimension;

1304. The sum of the above-mentioned score value vectors for the clicked documents in different search histories is the dynamic interest model of the user. The user's dynamic interest model can be expressed as: the user's search click history corresponding to the interest model R2 = dl + d2 + d3 + ... di, where di = (tl, t2, t3, ..., tj), di For the user to click on the interest model vector corresponding to a certain document i, when the user clicks on the document i, tj is equal to the sum of the word frequencies of all the words belonging to the jth interest dimension in the document i, i, j is a natural number.

When the user personalized data is the search history, in a specific example, the user's interest model may be further modified according to the specific situation, including:

When the user evaluates the document being clicked, the rating value vector for the document is forward weighted.

This step can be expressed as: If the evaluation of the document i is good, the di vector is multiplied by a positive constant c to indicate the importance of the document. di = c*di = (c*ti, c*t2, c *t3 , ..., c*tn). Or,

When the user does not evaluate the clicked document, the weighted value vector for the document is inversely weighted; the step can be expressed as: if the evaluation is not good for the clicked document i, di vector Multiplying the reciprocal of a positive constant c indicates that the importance of the document is reduced by di = l/c*di = (l/c*ti, l/c*t2, l/c*t3, ..., l/c*tn).

The score value vector for the document is decremented according to the time elapsed after the document was clicked.

For example, after a period of time, the value of tj is automatically reduced by a certain percentage, indicating that its importance decreases with time, until the value of tj is reduced to zero after a long period of time, then di is searched from the user. Deleted in history. For example, after every month after clicking, the value of tj is reduced by 10%. After a click for 10 consecutive months without a new click, the value of tj is reduced to zero. In the implementation process, the user personalized data may include both the static user profile of the user and the search history of the user. In this case, the obtained user's interest model may be referred to as the user's comprehensive interest model. Then, the method for extracting the user interest model may further include:

Normalizing the static interest model and the dynamic interest model, and combining the static interest model with the dynamic interest model as a comprehensive interest model; or

The static interest model, the dynamic interest model are first weighted and added, and the summed sum is normalized, and the normalized result is a comprehensive interest model. The specific process of selecting the member engine according to the search request, the user's interest model, and the meta index in the foregoing steps 804, 905 or 1005 will be specifically described below.

Referring to Figure 14, the selection process specifically includes:

1401. Acquire a first similarity between a first document in a database and a search request vector Q, where the first document satisfies a condition with the highest similarity with the search request vector Q; wherein, the search request vector 2 = (≠^-≠- ≠ 1, ^ is the weight of the term in the search request.

1402: Obtain a second similarity between the second document in the database and the user's interest model R, and the second document satisfies the matching threshold with the vector Q′ (ql′, q2′ . . . qm′). Based on the condition that the user's interest model vector R has the highest similarity, the vector Q' is a conversion form of the search request vector Q for the user's interest model vector R, and the user's interest model vector? = ir\, r2,. .ri...., rn), the weighting score for the i-th dimension of the user's interest model. 1403. Select a larger one of the first similarity and the second similarity as the similarity between the combination of the search request and the user's interest model and the database.

1404. Repeatly adopting the foregoing method to obtain a similarity between the combination of the search request and the user's interest model and each database, where each database corresponds to a member engine;

1405. Sort each database according to the similarity between the search request and the user's interest model and each database, and select a member engine corresponding to one or more databases with a similar degree of similarity. Specifically, the process of obtaining the first similarity and the second similarity in the above 1401~1402 may include different steps (that is, adopt different formulas or algorithms), and several specific examples are described below.

In one case: qj * giajj * anwj) The 1401a steps include: Calculation ( ₍ qi * gidfi * mnwi _{+ ≠i /|Q|} +

^rj* anvj

The value of j = ^l /|R|), where IQI is the modulus of the search request vector Q, and |R| is the modulus of the user's interest model R; this value is taken as the first similarity.

Max

The 1402a step includes: calculating l ≤ ≤ "(if _(s i _m(V (picture ^ν , ^β " ^ν /' ^≠ , ^{1 ≤} · / ' ^≤ ")) , Q ') > T) then

* anvj ^qi* gidfi* anwi

((fi*mnvi + , ) /| _R |+ ≡ί /|Q| ), the value is taken as the second similarity; wherein Q' is calculated as: if the term ti belongs to the user's interest model The range of a dimension, the value of qi is mapped to the weight of the dimension of the user's interest model, and then the weights of the same dimension are added to obtain qi', and then normalized; V is ^^ and ^^^ , ^^^ consists of vectors; _{iim{ Y{} mnvi, anvj{j≠i,\<j<n) ₎ , _Q ,) is the cousine similarity of the vector _v and the vector Q'; T is a threshold, and 0<T≤ 1; i, k, j, n are natural numbers. In another case: qj * giajj * anwj) The 1401b step consists of: Calculating 1 ≤ A ( _{ qi * gidfi * mnwi _{+ ≠i} /|Q| +

The value of ^rj*anvj*IM _gidjj / ), where |Q| is the modulus of the search request vector Q, |R| is the modulus of the user's interest model R; this value is used as the first similarity.

Max

The 1402b step includes: calculating 1 ≤ ≤" (if (sim( V( IM_gidfi*mnvi , IM _gidjj * anvj(j≠ i,\ < j≤n) ) , Q')> T ) „ qi * gidfi * anwi) then( ( ri * mnvi * IM _ gidfi + ^j* anvj * IM _ gidff ) /|R + '. ⁼¹ /|Q ), the value as the second similarity .

Wherein, the calculation method of Q' is: If the term ti belongs to a range of a dimension of the user's interest model, the value of qi is mapped to the weight of the dimension of the user's interest model, and then the weights of the same dimension are added to obtain qi. ', then normalized; V is a vector consisting of IM_gidfi* mnvi and - gidj} * anvjij≠ i,l≤ j≤ n); sim( V ( IM _ gidfi * mnvi , IM _ gidff * anvj (J≠ i,\ < _/ ≤")) , Q' ) is the cousine similarity of vector V and vector Q'; τ is a threshold and 0 < Τ ≤ 1. Another case: gidjj * Anwj) The 1401c steps include:

a value of /|Q| + ), where |Q| is the modulus of the search request vector Q, and |R| is the modulus of the user's interest model R;

This value is taken as the first similarity.

Max

The 1402c step includes: calculating 1 ≤ ί' ≤ " ( if (sim( V ( mnvi,

≠ i,l≤j≤n)) , Q')> T ) k

„ qi * gidfi * anwi) then( ( ri * mnvi * IM _ gidfi + ^j* anvj * IM _ gidff ) /|R + '. ⁼¹ /|Q ), the value as the second similarity ;

Wherein Q' is calculated as: If the term ti belongs to a range of dimensions of the user's interest model, the value of qi is mapped to the weight of the dimension of the user's interest model, and then the weights of the same dimension are added to obtain qi' And then normalized; V is a vector consisting of mnvi and ^ ^V /≠U; ^ _{m{ N(} mnvi, anvjU≠iA<j<n)^ , Q,) is vector _v and vector Q' Cousine similarity; T is a threshold and 0 < T ≤ 1. Through the calculation and selection of the foregoing 1401~1405, the selected database of the member engine has a good similarity with the combination of the search request and the user's interest model, thereby improving the accuracy of the search and saving the resources of the search system. , improve the efficiency of the search.

Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be implemented by means of software plus a necessary general hardware platform, and of course, by hardware. Based on such understanding, the above-described technical solutions may be embodied in the form of software products in essence or in the form of software products, which may be stored in a computer readable storage medium such as ROM/RAM, magnetic Disc, CD, etc., including a number of instructions to make a computer device (available The method described in various parts of the various embodiments or embodiments is performed by a personal computer, server, or network device. In the foregoing embodiments, the selection of the member engine fully utilizes the rich user data, and performs personalized selection, so that the member engine that best meets the user's personalized interest needs can be selected to serve the user, and the member engine selects the scheduling. Accurate, thus achieving the purpose of accurate search.

The embodiments described above do not constitute a limitation on the scope of protection of the technical solutions. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the above-described embodiments are intended to be included within the scope of the technical solutions.

In another embodiment of the present invention, similar to the foregoing embodiments, the difference is that the meta index contains information: the global reverse document frequency of the term ti relative to the database can be replaced by gidfi=log(n/(gdfi) +l)), where gdfi is the sum of the number of documents containing the term ti in the database or subdatabase of all member engines, and n is the sum of the total number of documents included in all member engines;

The global reverse document frequency corresponding to the i-th dimension of the document's interest model can be replaced by IM _gidfi=log(n/(IM _gdfi+l)), and IM _gdfi is the database or sub-database corresponding to all member engines. The sum of the number of documents for the term of the i-th dimension of the interest model, n is the sum of the total number of documents included in all member engines. FIG. 15 is a schematic structural diagram of still another search system according to an embodiment of the present invention. The system is similar to the system described above with respect to Figure 1, and includes a search client 10, a search service subsystem 20D, and at least one member engine 301, 302 or 303. The search client 10 is configured to send a search request to the search service subsystem 20D; at least one member engine 301, 302 or 303 is configured to provide the search service subsystem 20D with a meta index of the member engine, and upon receiving the search service After the search request sent by the system 20D, the search is completed.

The search service subsystem 20D shown in Fig. 15 is composed of a search server 201D, an application server 204, and a user database 202D.

The user database 202D is used to store or provide personalized data for the user.

The application server 204 is configured to receive a search request sent by the search client, extract the interest model of the user from the user personalized data according to the search request, and send the search request and the user interest model to the search server 201D.

The search server 201D is configured to receive the search request sent by the application server 204 and the interest model of the user, and receive the meta index reported by each member engine 301, 302 or 303, and according to the meta index of each member engine 301, 302 or 303. And the search request and the interest model of the user select a member engine 301, 302 or 303; send the search request to the selected member engine 301, 302 or 303.

Referring to FIG. 16, a structural diagram of the search server 201D and the application server 204 in the search system architecture of FIG.

The application server 204 includes a search request receiving module 2041, configured to receive a search request sent by the search client. The interest model extraction module 2042 is configured to extract a user's interest model from the user's personalized data according to the search request. Similarly, the user's interest model may be extracted from the user's personalized data after receiving the search request, or the user's interest model may be extracted from the user's personalized data in advance, and the pre-extraction is directly obtained after receiving the search request. User's interest model. The sending module 2043 is configured to send the search request and the user's interest model to the search server 201D.

The search server 201D includes: a receiving module 201D1, configured to receive a search request sent by an application server, and a user Interest model. The meta-index collection module 201D2 is configured to receive a meta-index reported by each member engine. The member engine selection module 201D3 is configured to select a member engine according to the meta index of the respective member engines, the search request, and the interest model of the user. The search request distribution module 201D4 is configured to send the search request to the selected member engine, so that the selected member engine completes the search according to the search request.

Referring to Figure 17, a flowchart of the operation of the system shown in Figure 15 is shown. The search method includes:

1700. The search server receives a meta index of each member engine.

Specifically, each member engine may actively report its own meta index to the search server, or the search server may request the member engine's meta index to each member engine.

1701: The application server receives a search request sent by the search client.

1702. The application server extracts the user's interest model from the user's personalized data.

Specifically, the user's interest model may be extracted from the user's personalized data after receiving the search request, or the interest model may be extracted from the user's personalized data in advance, and after receiving the search request, directly extracting the pre-extracted user's Interest model.

1703. Send the user's interest model and the search request to the search server.

1704. Select a member engine according to the search request, the user's interest model, and the meta index.

Specifically, the similarity between the search request and the interest model and the database corresponding to the member engine is calculated according to the search request, the interest model, and the meta index, and the member engine with high similarity is selected. For specific methods of selection, reference may be made to the foregoing specific embodiments.

1705. Send the search request to the selected member engine to facilitate the search by the selected member engine. It will also be understood by those skilled in the art that in the above embodiments of the system and the server, some or all of the modules may be integrated or distributed. For example, the functionality of the search service subsystem can be composed of a search server, a dispatch server, or an application server, or it can be integrated on one server. User databases can exist independently or on any of the aforementioned servers. The different modules in the aforementioned search server, dispatch server or application server can also be integrated in any combination.

It should be noted that, in an implementation manner of the foregoing methods, systems, and servers, the steps of extracting the user's interest model from the user personalized data according to the search request may include different situations: After receiving the search request, extracting the user's interest model from the user personalized data; or extracting the user's interest model from the user personalized data in advance, and after receiving the search request, directly extracting the advance The user's interest model is extracted from the user personalized data.

It will be understood by those skilled in the art that the user's interest model in various embodiments of the present invention can be equivalently replaced with personalization data of other various types of users. The user's interest model may be an expression of the user's personalized data, and the scope of protection of the present invention is of course not limited to this form of expression. It can also be understood by those skilled in the art that in various embodiments of the present invention, member engines are selected according to the meta-index of each member engine, the search request, and the personalized data of the user. In actual application scenarios, some considerations may be added. The selection of the member engine, or further processing of the selected member engine, such as integration, filtering, etc., is then searched by the finalized member engine.

Based on the description of each of the above embodiments, it will be appreciated that the present invention includes the following specific embodiments:

A search server, the search server includes:

a search request receiving module, configured to receive a search request sent by the search client; The interest model extraction module extracts the user's interest model from the user personalized data according to the search request; the meta index collection module is configured to receive the meta index reported by each member engine;

a member engine selection module, configured to select a member engine according to the meta index of the respective member engines, the search request, and the interest model of the user; and a search request distribution module, configured to send the search request to the selected a member engine, such that the selected member engine completes the search based on the search request.

A search server, the search server includes:

a search request receiving module, configured to receive a search request sent by the search client, and send the search request to the member engine selection request sending module;

The interest model extraction module is configured to extract the user's interest model from the user personalized data according to the search request, and send the user's interest model to the member engine selection request sending module;

a member engine selection request sending module, configured to send the search request and the user's interest model to a scheduling server, so that the scheduling server is configured according to the meta index of the respective member engines, the search request, and the user's interest Model selection member engine;

a member engine selection result receiving module, configured to receive a selected member engine returned by the scheduling server; a search request distribution module, configured to send the search request to the selected member engine, to facilitate the selected member The engine completes the search based on the search request.

A scheduling server is operative to communicate with the foregoing search server, the scheduling server comprising:

a member engine selection request receiving module, configured to receive a search request sent by the search server and an interest model of the user; a meta index collection module, configured to receive a meta index reported by each member engine;

a member engine selection module, configured to select a member engine according to the meta index of the respective member engines, the search request, and the interest model of the user;

a member engine selection result returning module, configured to send the selected member engine to the search server, so that the search server sends the search request to the selected member engine, and the selected member engine is configured according to The search request completes the search.

A search server, the search server includes:

a member engine selection request sending module, configured to send the search request to the scheduling server, so that the scheduling server selects a member engine according to the meta index of the member engines, the search request, and the user's interest model;

A scheduling server, wherein the scheduling server communicates with the foregoing search server, including:

a member engine selection request receiving module, configured to receive a search request sent by the search server;

An interest model extraction module, configured to extract the user's interest model from the user personalized data according to the received search request, and Sending the user's interest model to the member's engine selection module;

a meta index collection module, configured to receive a meta index reported by each member engine;

a member engine selection module, configured to select a member engine according to the meta index of the respective member engines, the search request, and the user's interest model;

a member engine returning module, configured to send to the search server according to the selected member engine, so that the search server sends the search request to the selected member engine, the selected member engine according to the The search request completes the search.

A search server that includes:

a receiving module, configured to receive a search request sent by an application server and a user's interest model;

a member engine selection module, configured to select a member engine according to the meta index of the respective member engines, the search request, and the interest model of the user; and

A search request distribution module is operative to send the search request to the selected member engine to facilitate the selected member to complete the search based on the search request.

Claims

Rights request

What is claimed is: 1. A method of searching, the method comprising: • receiving a search request;

Extracting a user's interest model from the user personalized data according to the search request;

Get the meta index of each member engine;

Selecting a member engine according to the meta index of each member engine, a search request, and a user's interest model;

The search request is sent to the selected member engine to facilitate the selected member to complete the search.

2. The method according to claim 1, characterized in that

The step of receiving a search request, extracting a user's interest model, and transmitting the search request to the selected member engine is performed by a search server;

The steps of selecting a member of the engine include:

The search server sends the search request and the interest model of the user to a scheduling server;

The scheduling server selects a member engine according to a meta index of each member engine, the search request, and the interest model of the user;

The scheduling server sends the selected member engine to the search server.

3. The method according to claim 1, characterized in that

The step of receiving the search request and transmitting the search request to the selected member engine is performed by a search server; the step of extracting the user's interest model and the selecting member engine specifically includes:

The search server sends the search request to a scheduling server;

The scheduling server extracts a user's interest model from the user personalized data according to the search request, and selects a member engine according to the meta index of each member engine, the search request, and the user's interest model;

The scheduling server sends the selected member engine to the search server.

4. The method according to claim 1, characterized in that

Receiving the search request, and extracting the user's interest model from the user personalized data according to the search request is performed by the application server; and

The steps of selecting a member engine include:

The application server sends the search request and the user's interest model to a search server;

The search server selects a member engine based on a meta index of each member engine, the search request, and the user's interest model.

The method according to any one of claims 1 to 4, wherein the extracting the user's interest model from the user personalized data according to the search request comprises:

After receiving the search request, extracting the user's interest model from the user personalized data; or, extracting the user's interest model from the user personalized data in advance, and directly taking out the pre-received after receiving the search request The user's interest model is extracted from the user's personalized data.

The method according to any one of claims 1 to 4, wherein the user's interest model is

16

Correction page (Article 91) The interesting dimension represents the interest of the household; the score value is given for each interest dimension; the vector consists of the score values for each interest dimension; the user's interest model includes: a static interest cast and a dynamic interest model;

The static interest model is: obtaining a word frequency of a word belonging to a certain interest dimension in a static user profile of the user, calculating a sum of word frequencies of all words belonging to the interest dimension, as a score value of the interest dimension, each rating value Forming a score value vector, which is a user's interest model, referred to herein as a static interest model;

The dynamic interest model is: obtaining a word frequency of a word belonging to a certain interest dimension in a document that is clicked in the search history of the user, and calculating a sum of word frequencies of all words belonging to the interest dimension in the document, the sum of the word frequency As a score value of the document for the interest dimension, each score value of each interest dimension forms a score value vector for the document, and the sum of the score value vectors for different documents is the user's interest model, which is referred to herein as Dynamic interest model.

7. The method according to claim 6, wherein the method further comprises:

When the user evaluates the document that is clicked, the weighting of the rating value vector for the document is forward-weighted; or, when the user does not evaluate the document that is clicked, the rating for the document is The value vector is inverse weighted; or, the score value vector for the document is decremented based on the time elapsed after the document was clicked.

8. The method according to claim 6, wherein the step of extracting the user's interest model from the user personalized data further comprises:

Normally normalizing the static interest model and the dynamic interest model, and then calculating a sum of the static interest model and the dynamic interest model, and using the result as the user's interest model; or

The static interest model and the dynamic interest model are first weighted together, and the summed sum is normalized, and the result is used as the interest model.

9. The method according to claim 1, wherein the meta-index is statistical data about one or a combination of the following information: a database corresponding to the member engine, a sub-database, in the database An included document or record, a document or record contained in the sub-database, or a term contained in the document or the record.

10. The method according to claim 9, wherein the meta index comprises one of the following information or any combination thereof.

The term maximum normalized weight vector mnw=(mnwl , mnw2, ..., mnwi, ... mnwp), where mnwi is the maximum normalization of the term ti relative to all documents in the database or sub-database corresponding to the member engine Weights;

The term average normalized weight vector anw = (anwl , anw2, ..., anwi , anwp), where anwi is the average normalized weight of the term ti relative to all documents in the database or subdatabase corresponding to the member engine ;

The interest model of the document in the database or the sub-database has a maximum normalized weight vector mtw=( _mnV l, mnv2 , mnvi , mnvn ), where the ith dimension of the document's interest model is relative to the member engine The maximum normalized weight of all documents in the database or subdatabase;

The interest model of the document in the database or subdatabase averages the normalized weight vector anv = ( _an vl, anv2 , ......, anvi , ..., anvn), where anvi is the document's interest model The average normalized weight of the i-th dimension relative to all documents in the database or sub-database corresponding to the member engine;

The term ti is relative to the global reverse document frequency gidfi of the database, where gidfl = l/dfi, dfi is the number corresponding to the meta index

17

Correction page (Article 91) According to the number M of the document containing the term ti in the library _;

The global reverse document frequency IM-gidfi corresponding to the i-th dimension of the document's interest model, where IM_gidfi=l/IM_IDFi, IMJDFi is the term of the i-th dimension of the interest model belonging to the document contained in the database or sub-database The number of documents; the term ti relative to the global reverse document frequency of the database

Where gdfi is the sum of the number of documents containing the term ti in the database or subdatabase of all member engines, and n is the sum of the total number of documents included in all member engines; or

The global reverse document frequency corresponding to the i-th dimension of the document's interest model IM-gidfi=log(n/(lM-gdfi+1)), IM-gdfi is the database or sub-database corresponding to all member engines. The sum of the number of documents for the term of the i-th dimension of the interest model, n is the sum of the total number of documents included in all member engines.

The method according to claim 9 or 10, wherein the step of selecting a member engine according to the meta index of each member engine, the search request, and the interest model of the user comprises:

Obtaining a first similarity between the first document in a database and the search request vector Q, the first document satisfies a condition with the highest similarity with the search request vector Q; wherein, the search request vector ^: ^, ^ ² , '^ ...^^, ^ is the weight of the term "in the search request;

Obtaining a second similarity between the second document in the database and the interest model R of the user, and the second document satisfies the requirement that the matching degree with the vector Q' (ql ', q2' ... qm') satisfies the requirement The threshold is based on the condition that the user's interest model vector R has the highest similarity, wherein the vector Q' is a conversion form of the search request vector Q for the user's interest model vector R, and the user's interest model vector R = (rl, r2, ..ri ...., rn), _ri is the weight of the user's interest model j-th dimension heavy score;

Selecting a larger one of the first similarity and the second similarity as the similarity between the combination of the search request and the user's interest model and the database;

Repeating the above method to obtain a similarity between the combination of the search request and the user's interest model and each database, wherein each database corresponds to a member engine;

Each database is sorted according to the similarity between the search request and the user's interest model and the respective databases, and the member engines corresponding to one or more databases having the same similarity are selected.

12. The method according to claim 1, wherein the step of acquiring the first similarity comprises:

ΜαΧ . * . ,

> 3⁄4/ * giajj * anwj) rj ^ anvj

Calculating a value of l ≤ i ≤ k ( _{ qi * gidfi * mnwi _{+ J≠i / |Q|} + , / |R| ), the value being the first similarity;

The step of obtaining the second similarity includes:

Max

Calculate 1 ≤ / ≤ " ( _{if (sim(} γ( mnvi, anvj j≠ i,\≤_/≤") ) , Q,)> _T ) _THEN ( ₍ ri * mnvi + qi * gidfi * anwi

a value of /|Q| ) as the second similarity;

Where IQ1 is the modulus of the search request vector Q; |R| is the modulus of the user's interest model R; Q' is calculated as:

18

Correction page (Article 91) Ti belongs to a range of dimensions of the user's interest model, maps the value of qi to the weight of the dimension of the interest model of the fff household, and then adds the weights of the same dimension to obtain qi', and then performs normalization processing; a vector consisting of mnvi and ^FL " ^V ≠ ^ ^≤ J ^{≤ w} ); sun (

≠ " ^≤ J' ^{≤ n} ) ) , Q' ) is the cousine similarity of the vector V and the vector Q'; T is a threshold, and 0 < T ≤ 1; i, k, j, n are natural numbers.

13. The method according to claim 11, wherein: the step of acquiring the first similarity comprises: calculating a value of / _|R| );

The value is taken as the first similarity;

Max

Obtaining a second degree of similarity comprises: calculating i ^ w (if (sim ( V (IM - g id fi * Videos ^{^{1, IM - Sidfj * anvj {}} j ≠ i, \ <j≤n)), Q , )> _T )

« k

Υ rj * anvj * IM _ gidff V qi * gidfi * anwi) t _hen (( ^ri * ^mnvi * ^{IM -} S ^id fi + J-^ ) / |R| + '=' /|Q|) a value as the second similarity;

Where IQI is the modulus of the search request vector Q, |R| is the modulus of the user's interest model R; Q' is calculated as: If the term ti belongs to a range of dimensions of the user's interest model, the value of qi is mapped The weight of the dimension of the user's interest model, then add the weights of the same dimension to get qi', and then normalize; V is by IM-gidf^mnvi and IM_gidff *anvj-(j≠i,l ≤j≤n) a vector composed; sim( V( ^IM ― g ^id fi * ^mnvi , ^IM ― Sidj] * continued jU ≠ U ≤ _ ≤ ")), Q, ) is a vector _v and a vector Q, similar to _cousine Degree; T is a threshold, and 0 < Τ ≤ 1; i, k, j, n are natural numbers.

14. The method according to claim n, wherein the step of acquiring the first similarity comprises: calculating

^ΜθΧ qj* gidff *anwj) Y rj * anvj * IM _ gidff

l ≤ i ≤ k (

/| _Q | ₊ ;=· /|R| ); the value is the first similarity;

Max

The step of obtaining the second similarity includes: calculating ^{1 ≤} f ^≤ " ( if(sim( V(mnvi nvjU≠ i,\≤·/≤"))),

.* gidfi* anwi)

Q')>T) then( ( ^{ri /} " ^1M ― +

a value of /|Q|) as the second similarity;

Where IQI is the modulus of the search request vector Q; |R| is the modulus of the user's interest model R; Q' is calculated as: If the term ti belongs to a range of dimensions of the user's interest model, the value of qi is mapped The weight of the dimension into the user's interest model, then

19

Correction page (Article 91) The weights of the same dimension are added to obtain qi', and then normalized; V is a vector composed of mnvi and ^αην · "≠ ^ ^≤ J ^{≤ η} ); sim( V( ^{≤ ≤ n} ) ) , Q' ) Is the cousine similarity of the vector V and the vector Q'; T is a threshold, and 0 < T ≤ 1; i, k, j, n are natural numbers.

15. A system for searching, characterized in that the system comprises:

a search service subsystem, configured to receive a search request, receive a meta index reported by each member engine, select a member engine according to the meta index of each member engine, the search request, and a user's interest model, and send the search request to The selected member engine;

16. The system for searching according to claim 15, wherein the search service subsystem is further configured to extract a user's interest model from the user personalized data according to the search request, so as to facilitate The member index of each member's meta-index, the search request, and the user's interest model are selected.

17. The search system according to claim 16, wherein the search service subsystem comprises a search server and a user database;

The user database is configured to store or provide personalized data of the user;

The search server is configured to receive a search request sent by the search client, extract a user's interest model from the user personalized data according to the search request, obtain a meta index of each member engine, and obtain a meta index according to each member engine. And the search request and the interest model of the user select a member engine; send the search request to the selected member engine.

18. The search system according to claim 16, wherein the search service subsystem comprises: a search server, a dispatch server, and a user database, where

The search server is configured to receive a search request sent by a search client, extract a user's interest model from the user personalized data according to the search request, and send the user's interest model and the search request to the schedule a server, receiving a selected member engine returned by the scheduling server, and sending the search request to the selected member engine;

The scheduling server is configured to receive a user's interest model and the search request sent by the search server, and obtain a meta index of each member engine; according to the meta index of the member engines, the search request, and the user The interest model selects a member engine; returns the selected member engine to the search server;

19. The search system according to claim 16, wherein the search service subsystem comprises a search server, a dispatch server, and a user database, wherein

The search server is configured to receive a search request sent by a search client, send the search request to the scheduling server, receive a selected member engine returned by the scheduling server, and send the search request to the The selected member engine is configured to receive the search request sent by the search server, extract the user's interest model from the user personalized data according to the search request, and obtain a meta index of each member engine; According to the meta index of each member engine,

20

Correction page (Article 91) The search request and the user's interest model select a member engine; return the selected member engine to the search server.

20. The search system according to claim 16, wherein the search service subsystem comprises a search server, an application server, and a user database:

The user database for storing or providing personalized data of the user;

The application server is configured to receive a search request sent by the client, extract the user's interest model from the user personalized data according to the search request, and send the search request and the user interest model to the search server;

The search server is configured to receive a search request sent by an application server and a user's interest model, receive a meta index reported by each member engine, and perform a meta index, a search request, and an interest model of the user according to the member engine Select member engine: Send the search request to the selected member engine.

twenty one

Correction page (Article 91)